Four added cardiac phase harmonics reduced mean SEEG test skill for a fixed ECG model in nine adult rest records

We tested whether cardiac phase improves later SEEG voltage prediction beyond a fixed lagged ECG model. The public source was OpenNeuro ds003688, version 1.0.7. We used nine adult rest records and all 578 good SEEG contacts from an earlier AutoSci package. Both ridge models fit an early block and predict a separate later block. The added model had four sine and cosine phase harmonics. Mean test skill fell by 1.350 percentage points. The 95 percent adult bootstrap interval was -2.492 to -0.579 points. Skill fell in eight adults and rose by 0.020 points in one. All six planned checks had a negative mean change. Timing controls also tended to reduce skill. An added descriptive check found better training fit in all nine adults. The full numerical analysis passed a repeat check. Fresh public ranges rebuilt all nine input arrays exactly. The package preserves exact inputs, source versions, code, models, errors, draws, figures, and retrieval instructions. This result applies to the fixed model and selected short rest blocks. It does not rule out neural responses at heartbeats or prove a method for cardiac artifact removal.

SEEG · ECG · cardiac phase · heartbeats · ridge regression · prediction · model transfer · intracranial recordings